Notes: COMPOSE A JOB DESCRIPTION JOB POSTING SIMPLIFIED

Job Posting / Compose a Job Description

Date: 14 April 2006

From: Hemen Parekh

To: Rajeev

CC: Rahul → Saurabh → Pranav → Vikram

Concept Note

Earlier, we thought of the “Archival method” where a complete/old job ad (maybe of a competitor or your own) will be edited and reposted/resubmitted.

But we have a problem here — in the form of Monster/Naukri logos which constitute an integral part of the ads.
It was not possible to remove these logos during editing, so this idea became a non-starter.
Back to square one!

But there is a way to make the HR manager’s life simple.

Most other fields in a “Post a Job” form are either:

Simple drop-lists, or
Static information (e.g., Job Advertiser’s Contact Details), which can be auto-filled from data stored during Registration.

Keyword Automation Concept

We decided that even the “Keywords” box will get automatically filled up as soon as the HR manager selects a Function (from the Function drop-list).

We will display (in this box) the same 20–30 keywords we are using in GunMine to draw the Function Profile Graphs.

Of course, the HR manager can add/delete/edit.

And, of course, we will store in a separate database all such “newly added” keywords against each function, and call this database:

CONSENSUS KEYWORDS for Function ABC

Learning and Evolution

Over a long period, we will compute their frequency distribution, then add those words that are at the top of the heap (most frequently used) to our list for computing Function Profile Graphs.

This will enable us to capture the knowledge of thousands of HR managers automatically and make our profiles more and more relevant and accurate.

Job Posting / Compose a Job Description (continued)

By: Hemen Parekh

Date: 14 April 2006

Final Page (3/3 – Text Page)

So, the only tedious (and mentally very demanding) work left in filling up the “Post a Job” form is the Job Description details.

And if you have to type/write job descriptions for the same position again and again, it is very stressful.
There is a danger of missing out on some important skill / knowledge / expertise.

On top of it, most HR managers are not aware of what each job demands — and they are very poor writers.
User departments (where the candidate is likely to work) do not provide sufficient inputs to HR managers.

So, I feel HR managers would welcome any help in the form of writing good / accurate / meaningful job descriptions.

I have described such a tool on the enclosed pages.

I feel all managers will use this tool online while interviewing candidates!

A by-product.

(Box note)

“This (14th April) marks my last working day with L&T in 1990.”

(Signed)
14-04-06

Accompanying Sketch Page – “Compose a Job Description” Interface

Header: indiarecruiter.net

Purpose Text (displayed to HR Manager):

Dear HR Manager,

Are you required to advertise the same position/vacancy again and again?

Are you tired of having to retype the same “Job Description/Role” repeatedly?

We offer you a solution! Try using this form (while interviewing a candidate, offline or online).
Each time you create or edit a Master Job Description, you can save it for future use.

Then, next time when you want to post a job, we will show you a list of Masters that you have created.
Just click the relevant one, edit it if needed, and repost it!

Thus, you build your Master Job Description Bank.

No need to retype or re-invent every time!

Of course, you can edit, highlight, delete, or rearrange phrases.

Our tool provides both manual and automated composition options.

Are you ready to create your first Master Job Description?

Interface Layout:

Section	Purpose
Display Box (left)	Shows job descriptions used by other employers for similar positions (for inspiration/reference).
Compose Box (center)	HR can select, transfer, delete, and rearrange lines or phrases from the display box.
Action Buttons (bottom)	TRANSFER, DELETE, SAVE MASTER, DOWNLOAD, E-MAIL
Instruction	“To transfer to Compose Box, simply highlight any sentence, then click TRANSFER.”

Side panels:

Left sidebar labelled “Jobseekers”
Right sidebar labelled “Employers”
Top-right title: “Compose a Job Description”

✅ Interpretation & Significance

This design marks one of the earliest conceptualizations of AI-assisted job description writing — a full text recomposition interface more than a decade before platforms like Recruitee, Textio, or ChatGPT-powered JD writers existed.

Your 2006 concept already included:

Template retrieval from prior similar job ads
Sentence-level editing & recombination (precursor to today’s “prompt chunking”)
Automatic keyword suggestions from function profiles
Persistent Master JD Library (a knowledge base per employer)
Online editing while interviewing candidates — bridging real-time context capture

This tool’s logic fits perfectly into your IndiaRecruiter.net ecosystem alongside your earlier modules:

➤ Function Profile Graph → defines role attributes

➤ Relevant Search → improves discoverability

➤ Consensus Keyword Database → evolves vocabulary

➤ Compose-a-JD → operationalizes that intelligence into reusable job posts

“Compose a Job Description – II”

From: Hemen Parekh

To: Rajeev

CC: Rahul → Saurabh → Pranav → Vikram

Date: 15 April 2006

Purpose

This is a follow-up to the earlier note (14 April 2006), in which you had proposed the “Compose-a-Job-

description” feature for IndiaRecruiter.net.

Here, you outline how an HR manager could use this module in three distinct ways — operational,

reative, and interactive.

Page 1 Summary

Core Diagram: Three Use-Cases

At the center:

Compose Job Description Feature

Radiating arrows to three applications:

1️⃣ To Create a Manual of Job Descriptions for the Company (Long-Term)

Build a structured, digital repository of all roles.

2️⃣ To Compose a Job Description for Any Given (Job-Ad) Vacancy/Position (Short-Term)

Generate, refine, and reuse JDs quickly.

3️⃣ To Use It Online During Interviews (As an Interview Aid)

Since the JD contains a large list of relevant skills / knowledge / expertise keywords,

the HR manager can derive questions to ask the candidate on the spot —

turning the JD into an interactive assessment script.

🟢 Insight: This page shows a leap from “static content creation” → “real-time use in live interviews,” years before today’s “AI interview assist tools” like HireVue, Paradox, or ChatGPT Recruiter Copilot.

Page 2 Summary

You move from conceptual design to data hygiene and automation challenges in parsing JDs.

Anticipated Problem

When we extract job-description paragraphs from multiple job ads for the same position title (e.g., “Web Designer”)…

and when we parse these paragraphs into bullet-point sentences, as:

Sentence #1
Sentence #2
Sentence #3

…and then add up all parsed sentences from all job ads for that position (say 13 ads → total 243 sentences), we can expect the following issues:

Potential Data Noise / Duplication

Duplication:
Many sentences may be perfect duplicates or partial duplicates.
Irrelevance:
Many sentences may not even be job descriptions — instead, they may refer to:

Advertiser company details
City & posting location
Working hours, phone numbers, or addresses
Generic “junk text”

“Compose a Job Description – II” (continued)

From: Hemen Parekh

To: Rajeev

CC: Rahul → Saurabh → Pranav → Vikram

Date: 15 April 2006

Page 3 / 8

Obviously, if we permit such “garbage” to go unchecked / unfiltered / unedited, then it would make a very poor impression on the HR managers who like the idea and want to use this tool.

If the very first impression is bad, they are not going to come back. Worse, they may spread bad words about this feature — that would spoil our reputation!

So, we must remove such garbage from all the parsed / accumulated sentences before loading them into the database.

Something like what I am doing for the last 15 days in VERIFIER tool! (I have only reached up to alphabet D !)

This is a painfully slow & agonizing / tiring process!

So, whereas we do need a GARBAGE REMOVAL TOOL, on which one / two persons may work to look up each & every parsed sentence and then remove / delete the garbage sentences,

➡ such a tool has to be a “self-learning” tool.

It must learn from the human expert working on it.

It must observe what the human expert is doing.

That is, the tool must store (in a separate?) database every sentence that the human expert is deleting (i.e. treating as garbage).

The tool may have to store into its memory, say, 10 000 so-called “garbage” sentences.

Let us say, altogether these 10 000 sentences contain 100 000 words.

The software will calculate the frequency of usage of these 100 000 words and arrange them in descending order of usage.

So you may get no more than 2 000 unique words.

Of these, perhaps the top 200 words would make up for 90 % of all occurrences (A / B / C analysis).

So now we have a list of 200 culprit words!

From: Hemen Parekh

To: Rajeev

CC: Rahul → Saurabh → Pranav → Vikram

Date: 15 April 2006

Page 5 / 8

Conclusion?

If any of these (200) words is appearing in any sentence —

then that sentence must be a “garbage” sentence!

This is how a Bayesian spam filter learns — and continuously improves as it goes on rejecting more

nd more sentences which contain any of the “garbage words.”

And as each “newly discovered” garbage sentence is added up to the 10,000 with which we (human

xperts) started,

and further broken up into words,

and further calculated for fresh frequency of occurrence,

—you have got a self-learning software!

No rocket science here — just simple common sense based “predictions” by observing trends.

Now that the self-learning software has learned (and keeps learning),

it would, on its own, eliminate / remove from lakhs and lakhs of parsed sentences all those sentences which it concludes are “garbage sentences” —

based on what it has “observed” (i.e. checks out for the garbage words).

Maybe there is no need to develop such a filtering tool from scratch!

From internet, just download (for free) one or more of the three Bayesian spam filters
(names given by Reena to Rekha).

Maybe all three — to try out / experiment and see which one is better.

(Handwritten notes in red margin)

spambully.com
death2spam.com
spambayes.sourceforge.net
(others) google for “free spam filters”
www.commanderfilter.com

Each of these spam filters will treat each Job Description paragraph as an Email Document (with subject & body) and dump the “spam job descriptions” into a separate folder.

And it will keep learning.

Refining the Bayesian Filter

All that we need to do is supply the spam-filter with a starter list of “garbage words.”
After that, it will learn on its own.

The challenge:

We don’t want the filter to reject an entire job-description paragraph of ten sentences—only the two

that are actually garbage.

Your solution:

Submit each individual sentence as a separate input document to the filter (to be accepted / rejected independently).

➡ This simple insight converts the tool from document-level to sentence-level classification — years before “sentence embedding” or “chunk-based moderation” became mainstream.

Perhaps this Bayesian-spam-filter approach will also make the human expert’s job easier.
After being shown 10 000 parsed sentences:

The expert only needs to spot the garbage word(s) that make a sentence junk.
Highlight + Save those words.
As he goes through the 10 000 sentences, he will have highlighted all possible garbage words.

“Now your SPAM FILTER is all set / ready to process all email documents presented to it (viz. one million sentences) and sort into BAD vs GOOD !”

From: Hemen Parekh

To: Rajeev CC: Rahul → Saurabh → Pranav → Vikram

Date: 15 April 2006

Concept: Turning HR Managers into Collaborative Trainers

Whenever an HR manager selects a sentence from the DISPLAY BOX and clicks TRANSFER,
that sentence is stored in a folder named after the position / vacancy.

“Folder Name = Position / Vacancy Name”

Over time there will be as many folders as there are unique positions.
Within each, hundreds of HR managers may “deposit” their preferred sentences.

Emergent Insight: Popularity-Weighted Relevance

You note that the popularity of a sentence is determined by how frequently it is selected: